Project Gestalt

A practical approach to working with AI

Problem First.
Prompt Second.

What do you need to understand, and what will that help you decide?

Start there. Choose a part of the work for AI, decide how you will check its answer, and keep track of what you do next.

Use the five-step cycle

Before opening an AI tool

I need to understand ___
so I can decide ___.

The cycle

Make your choices visible.

Use these steps within a research activity or another task. Return to earlier steps as your understanding changes.

  1. Frame

    What do I need to decide?

    State the problem, intended outcome and constraints. Say what observable decision or action would show that the work helped.

  2. Select

    What will AI contribute, and what stays with me?

    Choose the tool and its task. Predict the output and set your checks before seeing the answer. Retain responsibility for interpretation and decisions.

  3. Apply

    What did I ask, and what did the tool see?

    Keep the exact request, relevant inputs, product and visible model, and the response excerpt you used. Record failures as well as useful outputs.

  4. Evaluate

    Did the answer hold up? Did it help me decide?

    Check the output against your prediction. Verify a consequential claim against a source. Test the outcome separately, or record when you will be able to test it.

  5. Adapt

    What should change—or be retained?

    Revise the request or division of work if the output falls short. Revisit the question if a useful output leaves the problem unresolved. Keep supported decisions with their limits and choose your next action.

Declare AI use. Record what the tool contributed and what you did yourself, including any AI help with your log.

Seven choices you can make about AI use

Record only settings that mattered to the result. Use “not shown,” “not used” or “could not confirm” where appropriate.

Turn and context
Start fresh or continue a conversation whose history matters?
Model and effort
Which visible model or mode fits the task and time available?
Context I supply
Which passages, files or instructions does the tool need?
Reach outside
Can it search or use files, and which sources did it actually reach?
Output surface
Will you work in chat or an editable document?
Delegation depth
Which tasks will you hand over, and which decisions will you make?
Data and cost
What may you share, and what access, time or cost limits apply?

Applying the cycle to research

Explore the landscape.
Keep the evidence in view.

A landscape review helps you orient yourself, refine questions and prepare for research design and subsequent formal review. The AI-use cycle supports your work within each activity.

  1. Define landscape questions.
  2. Explore collections.
  3. Compile domain learnings.
  4. Visualize topics.
  5. Conduct focus-area analysis and hand off to subsequent review.

Treat generated themes, gaps, claims and references as provisional. You remain responsible for the focus, interpretation, verification and decision to proceed.

Illustrative example

From a suggested term
to a justified next step.

This is a prepared teaching scenario, not a captured AI response or a reported research finding.

Your question. Which classification terms from a digital-twin paper belong in your next search?

The AI contribution. You ask for the authors’ terms and a supporting passage for each. The response suggests three labels and says they are used across disciplines.

Your check. The paper’s table supports the labels in that study. It does not establish the wider claim.

Your decision. Retain the labels as candidate search terms and record the table location. Read a paper from a second discipline to compare vocabulary before making a broader claim.

Try it in your next research session.

Choose one activity. Keep your question, the request, one checked claim and the decision it changed in your scientific diary or a fresh exploration log.

Get the editable log